Collocation: The Heart of Word Choice

Learning outcomes

You know the word “rain” and you know the word “strong.” You know “argument” and you know “heavy.” Each word is correct, each definition is right, and yet “strong rain” and “a heavy argument” are both wrong in a way no dictionary entry will warn you about. The problem is not your vocabulary. It is that English has already decided which words keep company with which, and that decision is mostly arbitrary. This page is about that hidden layer of word choice, the layer where most translated-sounding English actually goes wrong.

After studying this page, you can:

• Define collocation as words that co-occur far more often than chance, and place any word pair on the cline from free combination through collocation to fixed idiom.
• Tell a strong collocation from a weak one, and explain why strong ones must be stored whole.
• Choose correctly among the delexical verbs make, do, take, have, and give with their typical noun partners, the split that trips nearly every learner.
• Recognize the main collocation patterns (adjective plus noun, verb plus noun, adverb plus adjective) and produce the natural partner instead of a plausible substitute.
• Check a doubtful pairing against a corpus or a collocations dictionary, and avoid calques, the wrong pairings carried over from another language.

Before we dive in

Here is a sentence a capable learner might write in a status update: “We did an important decision after a strong discussion about the heavy problem.” Read it slowly. Not one word is misspelled. Every word means what the writer intends. A grammar checker passes it without complaint. And yet a native reader winces at every second word, because the sentence reaches for the wrong partner each time. The natural version is “We made an important decision after a heated discussion about the serious problem.” The meanings barely changed. The partnerships did.

This is the trap that survives long after grammar and vocabulary are solid. You can know ten thousand words and still pair them wrong, because knowing a word is not the same as knowing the words it travels with. English, like every language, has a vast web of habitual partnerships: “rain” goes with “heavy,” “coffee” goes with “strong,” “a decision” goes with “make,” “a deadline” goes with “meet.” These partnerships are called collocations, from a Latin root meaning “placed together.” A collocation is two or more words that co-occur far more often than chance would predict, the habitual company a word keeps.

The reason collocation matters so much is that most of these partnerships are arbitrary. There is no rule that explains why rain is “heavy” but coffee is “strong,” when both just mean “a lot of.” A learner cannot derive the right partner from meaning; you can only know it because you have met it before, stored as a chunk. This is the idiom principle seen up close: collocations are one of the main kinds of ready-made chunk, and getting them right is most of what makes the difference between English that is correct and English that is natural. This page maps the territory: what counts as a collocation, how tightly words can be bound, the verbs that cause the most trouble, and how to check a pairing you are unsure of.

The cline from free to fixed

Not every pair of words that appears together is a collocation. “A fast car” is just two words that happen to fit; you could as easily say “a red car,” “a new car,” “a cheap car,” and nothing about “car” pulls “fast” toward it. At the other extreme, “kick the bucket” (meaning to die) is so frozen that you cannot change a single word: “kick the pail” means nothing. Collocation sits between these two poles. Understanding word choice means seeing the whole range, because where a pairing sits on it tells you how much freedom you have.

Picture a line running from total freedom to total fixedness. At the free end, words combine by meaning alone and any sensible substitute works. In the middle, words still combine meaningfully, but the language strongly prefers certain partners over equally logical ones: this is the zone of collocation. At the fixed end, the words have fused into a single unit whose meaning you cannot read off the parts: this is idiom. The diagram below lays the cline out with an example at each stage.

Read the cline from left to right as a loss of freedom. With “a fast car,” you are in open territory: swap “fast” for “old” or “blue” and you have simply described a different car. With “heavy rain,” your freedom is already gone: “strong rain” and “big rain” are both wrong, even though they would be perfectly logical, because English settled on “heavy” and nothing else. With “strong coffee,” notice that the very same idea of intensity now demands a different word: not “heavy coffee,” but “strong.” And with “kick the bucket,” there is no freedom at all, and no logic either, since the meaning has nothing to do with buckets. The practical lesson is that the middle zone, collocation, is the dangerous one, because the words still make sense individually, so nothing warns you that you have picked the wrong partner.

Strong and weak collocation

The collocation zone is itself a spectrum, not a single setting. Some words are loyal to very few partners; others keep wider company. This difference between strong and weak collocation decides how much you have to memorize and how much you can reason out.

A strong collocation is one where a word allows only a tiny set of partners, often just one. “Rancid” goes with “butter” and very little else; you can have rancid butter but not rancid bread (that is “stale”) or rancid milk (that is “sour”). “Foot the bill,” “shrug your shoulders,” “blink your eyes”: each verb is almost wholly committed to its noun. Strong collocations behave like near-idioms; you have to store them whole, because no general rule will hand you the partner.

A weak collocation is one where a word combines freely with many partners, each still natural. “A nice day,” “a nice meal,” “a nice person,” “a nice surprise”: “nice” attaches to almost anything pleasant, so it is weakly collocating and gives you little trouble. The weaker the collocation, the more it behaves like a free combination, and the less you need to memorize specific pairs.

The point of the distinction is to aim your effort. You do not need to study weak collocations; they take care of themselves. You need to hunt for and store the strong ones, because they are the ones that will catch you out, and the ones where a wrong guess sounds most foreign. The table earlier in your study of chunks lives here too: when you meet a noun, the question worth asking is which verbs and adjectives are strongly committed to it.

The delexical verbs: make, do, take, have, give

Nowhere does collocation trip learners more than in a small set of very common verbs: make, do, take, have, give. These are called delexical verbs, because in these uses they have been almost emptied of their own meaning (“de-lexical” means the lexical content has been stripped out). The meaning lives in the noun, and the verb is just the conventional partner that carries it. “Make a decision” means little more than “decide”; the verb “make” adds almost nothing except that it is the verb English happens to use with “decision.”

Because the verb carries no meaning, you cannot reason your way to the right one. “Make a decision” and “do a decision” are equally logical, since “do” is the generic verb for performing an action and a decision is something you perform. Yet only “make a decision” is English. The choice is pure convention, learned pair by pair. This is why “do a mistake” (instead of “make a mistake”) and “make your homework” (instead of “do your homework”) are among the most common errors at every level, including advanced. The table below gives each delexical verb several of its typical noun partners, so you can see the patterns and the lack of them.

A few patterns hide in the table, but only a few, and they leak. “Make” leans toward producing something that did not exist before (a decision, a profit, a mess), while “do” leans toward carrying out an activity (the dishes, research, your homework). That tendency helps a little, but it cannot be trusted: you “make an effort” but “do harm,” and both are about producing an outcome. “Have” and “take” overlap confusingly: you “have a look” and “take a look” (both fine), but you “have a meeting,” not “take a meeting” (except in some business slang). The honest summary is that these are conventions to be memorized, lightly guided by tendency, and the table is a starting store, not a rule you can extend safely.

Now test the choice on a fresh case before moving on.

The main collocation patterns

Collocations are not random pairings scattered through the language; they fall into a handful of grammatical shapes. Knowing the shapes helps you hunt for the right partner, because you learn to ask “which adjective goes before this noun?” or “which verb goes before this noun?” rather than guessing blindly. Three patterns carry most of the weight.

The first is adjective plus noun. Here an adjective is committed to a particular noun for a particular shade of meaning. The English for intense rain is “heavy rain”; for intense coffee, “strong coffee”; for intense concern, “deep concern”; for a forceful argument, “a strong argument”; for a forceful disagreement, “a heated disagreement.” Notice that the underlying idea, intensity, is the same across all of them, but the adjective changes with the noun, and you cannot predict which. “Strong rain,” “deep coffee,” and “heavy concern” are all wrong, and all sound translated.

The second is verb plus noun, the pattern that dominates the workplace. A task has a conventional verb. You “raise a concern,” “meet a deadline,” “ship a feature,” “address an issue,” and “roll out a change.” Reach for a logical-but-wrong verb and the sentence turns foreign at once: “say a concern,” “reach a deadline,” “send a feature,” “solve an issue” (sometimes acceptable, often not the natural fit), “publish a change.” The delexical verbs above are a special, extreme case of this pattern, where the verb has lost its meaning entirely; ordinary verb plus noun collocations keep more of theirs but are still conventional.

The third is adverb plus adjective, the pattern that polishes register the most. Certain adverbs intensify certain adjectives and no others. You are “bitterly disappointed,” “highly unlikely,” “deeply concerned,” “fully aware,” “painfully obvious.” Swap the adverb and you get something that means the same but sounds wrong: “strongly disappointed,” “highly unbelievable” (should be “highly unlikely”), “deeply happy” (should be “deeply content” or just “very happy”). These pairings carry a lot of the fluency signal precisely because they are easy to get almost right and still betray you.

Seeing the wrong, the awkward, and the right side by side is the fastest way to feel the line, since the wrong versions are all grammatical and all defensible by meaning. The table sorts a set of common cases the way a native ear sorts them.

Read down the “awkward or wrong” column and notice that every entry is defensible. “Strong rain” uses an intensity word for an intense thing. “Say a concern” uses the verb for producing words about a thing made of words. Each error is logical, which is exactly why meaning cannot save you and why these pairings have to be stored, not derived. The “very disappointed” row makes a softer but important point: sometimes a substitute is not wrong, merely weaker, and choosing the marked collocation (“bitterly disappointed”) is the difference between adequate English and native English.

Why collocations are arbitrary, and how to check them

By now the central difficulty is clear: collocations are arbitrary, so you cannot reason your way to them. There is no deep reason rain is heavy and coffee is strong. Another language might pair them the other way, or use the same word for both, and many do. This arbitrariness is not a flaw in English; every language has its own web of partnerships, and every web is mostly accidental, the residue of how that language’s speakers happened to settle. The consequence is unavoidable: collocations must be stored as chunks, the way the idiom principle describes, and cannot be generated from rules.

This is also why the most dangerous source of collocation error is your own first language. When you do not know the English partner, the strong temptation is to translate the partner you would use at home. This produces a calque, a word-for-word carryover that is grammatical in English but uses your language’s partnership instead of English’s. “Strong rain” (from languages that intensify rain with their word for strong), “do a mistake” (from languages that use their “do” verb with mistakes), “say a lie” (from languages that pair their “say” verb with lies, where English insists on “tell a lie”): each is a calque, and each sounds exactly as foreign as it is. The whole family of these carried-over pairings is the subject of calques natives never say.

Since you cannot derive collocations and cannot trust translation, you have to look them up, and there are two reliable tools. The first is a collocations dictionary, a reference organized not by definitions but by partnerships: you look up “rain” and it lists the adjectives (heavy, light, torrential, steady) and verbs (rain falls, rain eases) that go with it. The second is a corpus, a large searchable collection of real text where you can see which partner actually appears and how often, which is the most direct evidence of all. Learning to read a corpus yourself, so you can settle any collocation question from the evidence, is covered in reading the corpus yourself.

To make the arbitrariness concrete, the chart below contrasts how often, in rough illustrative terms, “make” and “do” each pair with two nouns. It shows the pattern you would find in any large corpus: each verb dominates with one noun and barely appears with the other, with no logical reason for the split.

Read the four bars in pairs. For “decision,” the “make” bar towers and the “do” bar is a stub: “make a decision” is overwhelmingly the form, “do a decision” almost never occurs. For “the dishes,” the picture flips exactly: “do the dishes” dominates and “make the dishes” is rare. There is no meaning-based reason for the flip; a decision and the dishes are both things you carry out. The chart is illustrative rather than measured, but the shape it shows, near-total dominance of one verb per noun, is what a real corpus search returns and is why these pairs must be learned individually.

Register-bound collocations

One last layer separates competent collocation from expert collocation: many partnerships are tied to a register, a level of formality and a setting. Picking the right collocation is not only picking the natural partner, but picking the partner that fits the situation. The same idea often has a formal collocation and a casual one, and using the wrong register is its own kind of error, even when the pairing itself is correct.

Consider the idea of starting to act on a plan. In a casual setting you might “get the ball rolling” or “kick things off.” In a formal report you would “initiate the process” or “commence implementation.” All four are correct collocations; none is wrong in the abstract. But “get the ball rolling” in a board paper looks unprofessional, and “commence implementation” in a quick team chat sounds stiff and cold. The collocation carries its register with it, exactly as the idiom principle notes for chunks in general: when you store the pair, you store its setting too.

This matters most in professional English, where the wrong register reads as a misjudgment rather than a mistake. “Touch base,” “circle back,” and “loop someone in” are natural business collocations but belong to informal workplace talk; a formal email prefers “follow up,” “return to this,” and “include someone.” A learner who has collected collocations from casual sources and deploys them in formal writing, or the reverse, sounds off in a way that is hard to name but easy to hear. The remedy is to label every collocation you store with its register, the same way you label it with its meaning, so that recall hands you not just a natural partner but a situation-appropriate one.

Mental Model: store the company, not the word

Here is the reframe that turns all of this from a burden into a method, and it corrects the model almost every learner starts with. The wrong model is that vocabulary is a set of words, each with a meaning, and that fluent word choice means knowing more words and reaching for the most precise or impressive one. Under this model, when you want to express intensity you scan your store for an intensity word and attach it, which is exactly how “strong rain” and “deeply happy” get produced. The model fails because it treats each word as independent, when in reality words live in committed partnerships.

The better model is to store the company a word keeps, not the word alone. A word in your vocabulary should arrive with its partners attached: “rain” comes bundled with “heavy, light, torrential, pouring”; “decision” comes bundled with “make, reach, tough, snap, final”; “deadline” comes bundled with “meet, miss, tight, looming, extend.” You are not learning “rain” and “decision” as points; you are learning small constellations, the word at the center and its habitual partners around it. The diagram below shows one such constellation, a node word and its collocate cloud.

Read the diagram as the unit you should actually learn. The center is the noun, but the value is in the spokes: the verbs that act on a decision and the adjectives that describe one. When you meet “decision” in reading, the productive move is not to confirm you know what it means (you do) but to collect a spoke you did not have, “reverse a decision” or “a snap decision,” and add it to the cloud. Over time the cloud is what you reach for, so the natural partner arrives already attached and “do a decision” never even forms. This is the same redirection of effort the idiom principle asks for, applied at the level of single words: learn constellations, not points.

Common mistakes

The first mistake is the make and do confusion, treated above: “do a mistake,” “make the dishes,” “do a decision.” It survives into advanced English because there is no reliable rule, only a stored list, and learners keep trying to reason the choice instead of memorizing the pairs. The fix is to treat every common “make” and “do” collocation as a small idiom learned whole.

The second mistake is the calque: carrying over a partnership from your first language. “Strong rain,” “say a lie,” “make a party,” “open the light”: each is grammatical English using a foreign partnership. The danger is that calques feel completely correct, because they are correct in the language you are translating from. The only defense is to distrust translated partners and check the doubtful ones against a corpus, never assuming the partner is the same across languages.

The third mistake is reaching for a single fancy word instead of the natural pairing. A learner who wants to sound advanced replaces “a serious problem” with “a grave conundrum” or “make a decision” with “render a determination.” Occasionally these land; usually they sound stilted, because natural English prefers the common collocation to the rare synonym. Aim for the natural partner, not the impressive one.

The fourth mistake is ignoring register: collecting collocations from one setting and using them in another. “Touch base going forward” in a formal proposal, or “commence the undertaking” in a quick message, are both register errors even though each collocation is real. Store every collocation with its setting, and match the setting when you use it.

The fifth mistake is trusting a grammar checker to catch collocation errors. Because collocation errors are grammatically perfect, automated tools pass them silently. “We did an important decision” has no grammar fault; only a collocation sense, or a corpus check, flags it. Do not assume that a sentence which passes a checker sounds native; the most foreign-sounding errors are the ones that survive the check.

Mastery Questions